Author(s): Nisha

Abstract: This study aims to develop a reliable deep learning framework for detecting pneumonia from chest X-ray images using attention-enhanced convolutional neural networks. Sourced from the publicly accessible Kaggle repository, the dataset consists of 5,863 labelled greyscale chest X-rays classed as "Normal" or "Pneumonia." To improve diagnostic visibility, the data was normalised, resized to 224×224 pixels, and histogram equalised. To handle class imbalance and increase model generalisation, data augmentation methods including rotation, flipping, and brightness adjustment were used. Class distribution plots, picture intensity histograms, and feature correlation heatmaps were part of Exploratory Data Analysis (EDA), which helped to steer modelling choices. Implemented were four models: a baseline CNN, VGG16 with transfer learning, CNN with self-attention, and a hybrid ResNet50-CBAM. Performance was measured by means of Accuracy, Precision, Recall, and F1-Score. Results indicated gradual performance enhancement with model complexity and attention processes. With better feature localisation and diagnostic accuracy, the ResNet50-CBAM hybrid model reached the greatest accuracy of 98.41% and F1-Score of 0.98. This study emphasises the effectiveness of improving both classification accuracy and interpretability-important factors in medical image diagnosis systems-by merging residual networks with attention modules.

DOI: https://www.doi.org/10.33545/27076571.2025.v6.i1c.153

Pages: 187-194 | Views: 2212 | Downloads: 1539

Download Full Article: Click Here